The standard claim in every vendor deck is that leadership already agrees artificial intelligence matters, so the remaining problem is the tool. For that to be true, the people signing the cheques would have to be able to read a proposal, set a procurement standard and tell a pilot from a demonstration. Statistics Canada’s own numbers say otherwise. Among Canadian businesses that already use artificial intelligence, only 51.7% of the large ones have trained their executives to use it, and at firms of one to four people the figure is 15.6%.
Key facts
- Statistics Canada’s Canadian Survey on Business Conditions for the second quarter of 2026 reports that among artificial intelligence users, 51.7% of businesses with 100 or more employees trained executives to use it, against 15.6% of businesses with 1 to 4 employees.
- The same survey reports employee training at 68.1% of artificial intelligence users with 100 or more employees and 24.0% of those with 1 to 4 employees.
- A Bank of Canada staff analytical note published in June 2026, drawing on a December 2025 Business Leaders’ Pulse survey of 314 firms, reports 70% of business decision-makers using artificial intelligence personally in a typical work week against 8% reporting significant use in core operations.
- McKinsey’s global survey published 25 August 2026, covering 1,719 respondents across 97 countries, reports 6% of organizations capturing at least 5% of earnings before interest and taxes from artificial intelligence.
- The MIT NANDA report on enterprise generative artificial intelligence, covering January to June 2025, describes a funnel in which 60% of organizations investigated task-specific tools, 20% ran pilots and 5% implemented successfully.
The gap is between personal use and operational use
The sharpest number in Canadian data is not an adoption rate. It is a pair. The Bank of Canada found that 70% of business decision-makers use these tools personally in a typical week, while 8% report significant use in their firm’s core operations. Those two figures describe the same people. The executive who drafts a board summary with a chatbot on Tuesday runs a company that has changed nothing about how it quotes, schedules or bills.
That gap is not a technology problem and it is not reluctance. It is a translation problem. Personal use requires you to open a window and type. Operational use requires somebody to decide which process changes, who owns the output when it is wrong, what the fallback is, how it gets audited and what the thing costs at volume rather than at trial. Every one of those is a management decision, and none of them can be delegated to whoever is most enthusiastic.
The strongest version of the optimistic case
Here is the argument for doing nothing, made properly, because it deserves better than it usually gets. Tools are improving faster than any procurement cycle can track. A company that commits in September is committing to a capability that will be cheaper and better in March, and possibly bundled free into software it already licenses. Waiting has a real return, and most of the firms congratulating themselves on being early have bought a subscription and a slide deck.
The optimists are partly right. The MIT work covering the first half of 2025 found 60% of organizations investigating these tools, 20% piloting and 5% getting to a working implementation, which means the overwhelming majority of the money spent so far has bought experience rather than output. Anybody who sat out that cycle lost nothing but a year of anecdotes.
Where the argument breaks is the assumption that the waiting is free. What the early cycle actually produced, for the firms that finished anything, is not a tool advantage. It is a procurement muscle. They now know what a real reference customer sounds like, what a pilot should cost, how long an integration takes in their own systems and which vendor claims collapse under a specific question. That knowledge does not get cheaper in March. It only gets acquired by doing it once badly.
What to actually do with the next vendor call
Three things, and none of them require a budget line. Name the three workflows in your business where the hours actually concentrate, which is almost never the workflow anybody complains about most. Put one person’s name against each, with authority to say no. Then write down, before the call, what result would justify the spend and what evidence you will accept as proof, because a demonstration is designed by the vendor and a pilot is designed by you, and the entire difference between the two is who chose the data.
The McKinsey figure worth holding is the 6%. That is the share of organizations globally getting at least 5% of earnings out of this, against 44% who say they are scaling it. The distance between those two numbers is not the technology failing. It is what happens when scaling is defined as deployment rather than as a change in what the business earns. AI adoption measured by seats sold will always look better than adoption measured by margin.
So the prediction, with its condition attached. If the Statistics Canada executive training figure for firms under twenty employees has not moved above 25% by the second quarter of 2027, Alberta’s small and mid-market adoption rate will still be trailing the national average, and the explanation offered will be access to capital rather than the thing that actually decided it, which is that nobody in the room could read the contract. Alberta business has the tools already. It has not yet bought the fluency to buy them well.
